2D Statistical Lung Shape Analysis Using Chest Radiographs: Modelling and Segmentation

Area of Science:

• Medical Imaging
• Anatomical Variation Analysis
• Pulmonary Disease Diagnosis

Background:

• Accurate lung shape analysis and understanding anatomical variations are crucial in medical imaging.
• Normal lung shape variations can indicate healthy lungs, while abnormal variations may signal pulmonary diseases.
• Inter-patient lung shape variability presents challenges in medical image analysis and segmentation.

Purpose of the Study:

• To develop and compare two distinct lung shape models for analyzing inter-patient 2D lung shape variations.
• To evaluate the impact of different reference points in registration on lung shape variation estimation.
• To utilize these models within an active shape model (ASM) framework for accurate lung field segmentation.

Main Methods:

• Development of two unique 2D lung shape models with differing reference points for registration.
• Application of Principal Component Analysis (PCA) to model lung shape variations.
• Implementation of an Active Shape Model (ASM) technique for lung field segmentation.
• Validation using leave-one-out cross-validation.

Main Results:

• The developed lung shape models, combined with PCA, effectively captured over 95% of total lung shape variations using only the first 7 principal component (PC) modes.
• The ASM-based segmentation method achieved high accuracy, with average Dice Similarity Coefficients of 97.1% for the right lung and 96.1% for the left lung.
• The proposed segmentation method demonstrated superior stability and accuracy compared to existing model-based techniques for inter-patient lung field segmentation.

Conclusions:

• The proposed lung shape models provide a robust representation of inter-patient 2D lung shape variations.
• The integration of these models with ASM offers a highly accurate and stable approach for lung field segmentation in medical imaging.
• This methodology holds significant potential for improving the diagnosis and monitoring of pulmonary diseases through enhanced medical image analysis.